What is going to happen to Jon Snow, Cersei Lannister and Daenerys Targaryen? Even if George Martin can’t write that fast as we want, high technologies come to help.
We can’t take this anymore!
George Martin, an american science fiction writer is working slowly on the book serie called “ A Song of Ice and Fire” from 1991. Five of seven initially planned books are published up to date. Over 25 years hundreds of thousands of its fans from around the world could not take the pressure of not knowing what will happen to their favorite characters.
When HBO, an american cable and satellite television network created a show called “Games of Thrones” based on Martin books, the waiting army of tortured fans waiting for new release has grew up to millions. And a help came from unlike source, from the technology able to calculate the story.
Neuronet is a system able to study. That means if you wrote a program using a certain sequence of actions it will not only follow your orders but also remember its previous experience and define mistakes and study on them. Neuronet has more and more implementation areas: forecasting, data collection and analysis, optimization, text creation upon certain criterias.
In short, the technology is developed in order to help people solve sophisticated tasks. And if the fans of Seven Kindgdoms of Westeros wanted to know the future, they could get what they want.
How is it possible?
The programmer Zack Thoutt with LSTM technology (from the term long short-term memory) has created a program able to manage the task George Martin cannot handle.
LSTM is a successful modification of a neuronet with a recurring or returning sequence well known to mathematician. Following string of number can serve as an example of such sequence: 2, 4, 8, 32 where every next number is a multiplication of two previous numbers.
LSTM can remember information long enough. In this way at the end of the page the neuronet will not forget who is Jon Snow and will apply his usual language. This is how the style is recreated. Moreover the program will learn the logic of previous events, will find patterns and based on that all will build the rest of the story. This is how the story is created.
In order to provide neuronet with food for thoughts Zack Thoutt has uploaded all published books. Then the programmer set the name of the character. That was the beginning of the story. It even looks like George writing style: in the third person, from the perspective of one of the characters.
Of course the storytelling in the book made by LSTM is not perfect. The program sometimes makes mistakes and finds not the best expressions.
Zack Thoutt is not 100% happy with the result. And one of the reasons for that is the same as Martin’s readers have: it is written too little. The neuronet is learning on the material created by the author. And there were an opportunity to upload more stories, the text would be better. But also the style of George Martin is a big challenge for a neuronet of current generation: it has too many epithets, adjectives and descriptions.
Be careful, spoilers!
You can read the beginning of the story “Winds of winter”, five chapters written by the neuronet of Zach Thoutt.
If you avoid spoilers this paragraph is not for you 😉
By the way, it is not the only attempt to create a book on this way. On the 8th of July of 2016 another programmer Max Deutsch published a book about Harry Potter written by a neuronet.
Max Deutsch also used a LSTM neuronet, but for program training he has used only the first book of the serie. His main goal was to create an algorithm able to write qualitative, beautiful and independent stories would be nice to read. In the book about Harry Potter he managed to keep the initial style and rythmics and that’s already pretty good.
The interesting point is, who will be identified as authors of such books?…